Codex 0.167.0: Agent Command Center, Backend Switching and Vertex AI Setup

OpenAI shipped Codex CLI 0.167.0 on November 6, 2026. It adds unread markers to the agent command center, backend switching that keeps your drafts, local image previews and a guided Google Cloud Vertex AI setup.

These changes make Codex easier to run as a team of parallel agents. This guide shows how to update, how each feature works and how to fix the common snags.

What is new in Codex 0.167.0

Codex CLI is OpenAI’s open source coding agent for the terminal. Version 0.167.0 is a stable release listed on both the official Codex changelog and the GitHub releases page.

Here are the headline features from the release notes:

• Agent command center: run codex agents --no-daemon to open it. The command center and the session picker now mark sessions with unread results.
• Backend switching: the terminal session picker lets you move between connected backends, and your drafts stay in place.
• Local image previews: on supported terminals you can preview image attachments while composing. Images that Codex viewed can be opened from clickable paths in the transcript.
• New sign-in menu: API key and cloud provider options now sit under “Sign in another way”.
• Vertex AI wizard: enabling gcp_setup_wizard guides you through Google Cloud Vertex AI setup.
• System proxy: you can opt into macOS or Windows system proxy settings from /experimental.
• Patch details: you can expand patch details in the transcript while an approval prompt is open.

The last four releases at a glance

Codex ships fast. Several features in 0.167.0 build on work from the last few weeks, so it helps to see them together.

So the Vertex AI provider itself arrived in 0.165.0. What 0.167.0 adds is a friendlier, guided way to set it up.

How to update Codex CLI

Use the same method you used to install Codex. Mixing methods can leave two copies on your machine, which is a common cause of “my version did not change”.

1. Check your current version. Run codex --version in a terminal.
2. Update with npm. Run npm install -g @openai/codex@0.167.0 to pin this exact release, or npm install -g @openai/codex@latest to take the newest one.
3. Or update with Homebrew. Run brew upgrade --cask codex if you installed with brew install --cask codex.
4. Or rerun the install script. On macOS or Linux run curl -fsSL https://chatgpt.com/codex/install.sh | sh. On Windows, use the PowerShell script from the Codex README.
5. Confirm the update. Run codex --version again. It should print 0.167.0 or newer.
6. Restart open sessions. Close running Codex windows so the new binary loads.

Tip: If you are new to Codex, start with my Codex guide covering the CLI, app and pricing, then come back here for the new features.

How to use the agent command center

The agent command center is a full-screen view of all your Codex sessions. It first appeared as the codex agents dashboard in version 0.149.0, for searching, starting, opening, renaming and stopping tasks.

Opening it

The 0.167.0 notes give this command:

codex agents --no-daemon

The --no-daemon flag runs the view without relying on a background server. You can also run plain codex agents, or open it from inside a running session with the /agents command.

What you can do there

• Search: filter sessions by name or project.
• Start: launch a new agent task.
• Open: jump into a running session.
• Rename: give a session a clear title like “auth refactor”.
• Stop: end a session you no longer need.
• Group by directory: since 0.165.0, sessions group by folder, which helps when you use git worktrees.

The new unread markers

When you run four or five agents in parallel, the hard part is knowing which one needs you. Version 0.167.0 marks sessions with unread results in both the command center and the session picker.

Open the marked session, read the result and the marker clears. This works nicely with codex queue, which sends a follow-up message to an existing session from outside it.

codex queue --session my-refactor "Tests pass. Continue with the schema change."

Note: For more on running agents side by side, see Codex vs Claude Code and my notes on Codex usage limits, since parallel agents use your plan’s limits faster.

Switching backends without losing drafts

A backend is the server that runs your Codex sessions. Many people only have the local one, but you may also connect remote app servers, for example a build machine.

In 0.167.0 the terminal session picker lists your connected backends. You can switch between them, and any half-typed prompt you left in a draft is kept.

1. Open the session picker. Start codex and open the picker, or use codex resume.
2. Choose a backend. Pick the connected backend you want from the list.
3. Pick a session. Open an existing session or start a new one there.
4. Switch back any time. Your earlier drafts are still waiting.

A related fix also landed. Saved chats now stay visible after you change model providers, so moving from OpenAI to Vertex AI no longer hides older chats.

Local image previews

Codex can take screenshots and mockups as input. Before 0.167.0 you attached an image and hoped it was the right file.

Now, on supported terminals, you can preview local image attachments while composing. Images Codex has viewed can also be opened from clickable paths in the transcript.

Which terminals work

The 0.164.0 notes say image previews need a terminal that supports the Kitty or Sixel graphics protocols. Common examples include:

• Kitty protocol: Kitty, WezTerm and Ghostty.
• Sixel: terminals with Sixel turned on, such as recent Windows Terminal builds and some xterm setups.
• Others: the default macOS Terminal does not show inline graphics, so you will see paths instead.

Tip: If previews do not show, the attachment still works. Codex sends the image to the model either way. The preview is only for you.

How to set up Google Cloud Vertex AI

Since 0.165.0, Codex has a built-in Vertex AI provider named gcp. That lets companies that buy models through Google Cloud run Codex against their own project and billing.

Version 0.167.0 moves API key and cloud provider options under “Sign in another way” and adds the gcp_setup_wizard flag. The wizard supports Application Default Credentials (ADC) or a credential file.

Before you start

• A Google Cloud project with billing and the Vertex AI API turned on.
• Access to the models in that project, such as GPT-6.1 Terra, which 0.166.0 added to the Vertex AI catalog.
• The gcloud CLI, if you want to use ADC.

Setup steps

1. Create local credentials. Run gcloud auth application-default login and sign in with your Google account. This stores ADC on your machine.
2. Turn on the wizard. gcp_setup_wizard is a feature flag. Codex flags can be turned on with --enable <name> or as features.<name>=true, so try codex --enable gcp_setup_wizard.
3. Open the sign-in screen. Choose “Sign in another way”, then the cloud provider option for Google Cloud.
4. Pick a credential type. Choose ADC, or point the wizard to a service account credential file.
5. Enter project details. Give your project and region when asked.
6. Test it. Send a small prompt and confirm a reply comes back without an auth error.

If you prefer config files, the provider is selected with this line in ~/.codex/config.toml:

model_provider = "gcp"

Warning: Vertex AI usage is billed to your Google Cloud project, not your ChatGPT plan. Never commit a credential file to git, and set budget alerts in the Cloud console.

Other fixes worth knowing

The rest of the release is smaller, but a few items fix real daily annoyances.

• Failed resumes: if a chat cannot resume, you can read its transcript and get an editable prompt to continue in a new chat.
• Clear screen: Ctrl+L clears the display, and earlier messages stay available through Ctrl+T.
• Network approvals: prompts now close when their request ends, and parallel requests to the same endpoint prompt one at a time.
• Voice chats: long realtime V3 conversations save finished speech into chat history.
• Compaction: automatic context compaction now accounts for output from failed responses.
• Linux memory: musl builds use fewer allocator arenas, which lowers idle memory.
• Docs: managed network rules under [permissions.network] and prompt-user command rules are now documented.

If you write project rules for Codex, my AGENTS.md guide covers how to keep agents on track across sessions.

My Take

This is a quality of life release, not a big model update. But the unread markers solve the main pain of running agents in parallel: you stop checking each tab by hand.

The Vertex AI wizard matters more for teams. Many companies already pay Google Cloud and want one bill. A guided setup with ADC lowers the barrier a lot.

If you only use Codex with a ChatGPT plan on one machine, update anyway for the resume and Ctrl+L fixes. It is a quick, low risk upgrade.

Troubleshooting: Common Problems and Fixes

• Version still shows the old number: you likely have two installs. Run which codex (or where codex on Windows), remove the extra copy, and update with one method.
• codex agents shows nothing: start at least one session first, then reopen the command center. Try codex agents --no-daemon if a background server is stuck.
• No image preview: your terminal probably lacks Kitty or Sixel graphics. Switch terminal or rely on the clickable path.
• Vertex AI permission denied: check that the Vertex AI API is enabled and your account has a Vertex AI user role in that project.
• ADC not found: rerun gcloud auth application-default login, or give the wizard a credential file path.
• Network errors behind a company proxy: open /experimental and opt into system proxy settings on macOS or Windows. You can also run codex doctor --network-only.
• Chats vanished after changing provider: this was fixed in 0.167.0, so update first.

Frequently Asked Questions

When was Codex CLI 0.167.0 released?

It was released on November 6, 2026. It appears on the official Codex changelog and on the openai/codex GitHub releases page.

How do I open the agent command center?

Run codex agents --no-daemon as shown in the release notes, or plain codex agents. Inside a session you can use /agents.

What does backend switching do?

It lets you move between connected backends from the terminal session picker. Your drafts are kept when you switch.

Do I need a special terminal for image previews?

Yes, previews need a terminal with Kitty or Sixel graphics. Attachments still work in other terminals, you just will not see the preview.

Is Codex on Vertex AI billed through ChatGPT?

No. When you use the gcp provider, model usage goes through your Google Cloud project and its billing.

Is Codex CLI free?

The CLI is open source and free to install. Using it needs a ChatGPT plan that includes Codex, an API key, or a cloud provider such as Vertex AI or Amazon Bedrock.

Which models can I use with the Vertex AI provider?

The 0.166.0 notes added GPT-6.1 Terra to the Vertex AI catalog. Check the model list in your own Google Cloud project, since access can differ by region.

Sources

• GitHub: openai/codex release 0.167.0
• OpenAI Codex changelog
• GitHub: openai/codex release 0.165.0 (Vertex AI provider)
• GitHub: openai/codex release 0.149.0 (codex agents dashboard)
• GitHub: openai/codex README (install)
• Google Cloud: Set up Application Default Credentials